Impressionable Years: The Impact of Parental Political ... · PDF file1 Impressionable Years:...

38
1 Impressionable Years: The Impact of Parental Political Socialization and Education on Later Political participation Vani Murugesan Javier M. Rodriguez David O. Sears University of California, Los Angeles Abstract The Impressionable Years hypothesis proposes that salient events during adolescence and young adulthood are particularly important for political socialization. In accordance with this hypothesis, this paper uses a longitudinal dataset to examine whether the political environment of 17 and 18 years olds produce lasting effects on their political participation 10, 20, and 30 years later. While previous analyses have relied heavily on regression modeling, we use SEM to allow for latent variables and mediation processes. Parental socio-economic status and political socialization during adolescence as well as the experience of a college education work in concert to produce lasting political engagement. Parental political socialization has a lasting impact on offspring political engagement, which in turn predicts the offspring’s political participation over the next 30 years. Receiving a college education produces a more stable effect on political engagement over time than does parental socialization. The main assumptions of the Impressionable Years hypothesis can be extended to actual political behavior, therefore.

Transcript of Impressionable Years: The Impact of Parental Political ... · PDF file1 Impressionable Years:...

1

Impressionable Years: The Impact of Parental Political Socialization and

Education on Later Political participation

Vani Murugesan Javier M. Rodriguez

David O. Sears

University of California, Los Angeles

Abstract The Impressionable Years hypothesis proposes that salient events during adolescence and young adulthood are particularly important for political socialization. In accordance with this hypothesis, this paper uses a longitudinal dataset to examine whether the political environment of 17 and 18 years olds produce lasting effects on their political participation 10, 20, and 30 years later. While previous analyses have relied heavily on regression modeling, we use SEM to allow for latent variables and mediation processes. Parental socio-economic status and political socialization during adolescence as well as the experience of a college education work in concert to produce lasting political engagement. Parental political socialization has a lasting impact on offspring political engagement, which in turn predicts the offspring’s political participation over the next 30 years. Receiving a college education produces a more stable effect on political engagement over time than does parental socialization. The main assumptions of the Impressionable Years hypothesis can be extended to actual political behavior, therefore.

2

Electoral and non-electoral participation are at the core of political science research agendas. Levels and

types of political participation not only describe the political character of human nature but they also serve

to assess the quality and health of a democracy. Powerful independent predictors such as education,

political engagement, and parental political socialization continue to cast the foundations of the state of

the discipline. Less is known, however, about how these factors work together and ultimately lead to

participation.

For decades education is found to be a central correlate/determinant of participation (Converse,

1972; Hillygus, 2005). The most intuitive explanation for its effects is that it provides the necessary skills

to participate, a process that usually occurs in the critical period of late adolescence/early adulthood, best

known as the “impressionable years.” The implementation of these skills, which outlast the experience of

receiving a college education, suggests that there are important mediators between educational attainment

and later participation. These mediators, however, are usually not accounted for in explanatory models.

A wealth of research is focused on the direct association of educational level with political participation,

thus the conventional wisdom that education is a primary determinant of political participation, even

though the mechanism through which this impact operates still is unclear.

We propose that education’s main contribution to political participation is through the facilitation of

subjective political engagement, an engagement that implies the implementation of the skills acquired

through education. Acknowledging the limitations of regression analysis at studying mediation, we

implement SEM, a methodological alternative that permits us to examine the nature of the multivariate

interactions and latent structures that underlie the complexity of the processes that lead an individual to

participate and that are generally not taken into account. These structural models show that education

does not have a direct effect over political participation. Rather, both parental socialization during

childhood and the education an individual is exposed to during the impressionable years work in concert

3

to produce political engagement, which in adulthood leads to political participation. Significantly, the

main assumptions of the Impressionable Years hypothesis can be extended to actual political behavior.

Analysis of mediation processes that are usually not taken into account evince that attitudes and behaviors

are more related than much research has suggested.

Education

The correlation between educational attainment and higher levels of health, wealth, knowledge, prestige,

institutional and human resources, and ultimately of power, is perhaps one of the most reliable across

disciplines. Not surprisingly the benefits afforded by educational attainment are further transferred onto

the political sphere. Those individuals with higher levels of education have consistently shown higher

levels of both electoral and non-electoral participation in the United States (Converse, 1972; Leighley and

Nagles, 1992; Miller and Shanks, 1996; Nie, Junn, and Stehlik-Barry, 1996; Niemi and Junn, 2005;

Rosenstone and Hansen, 1993; Verba and Nie, 1972; Verba, Schlozman, and Brady, 1995; Wolfinger and

Rosenstone, 1980).

There are abundant possible reasons for why education has such a reliable, positive impact on

participation. A college education, for example, has been defined as an experience that both educates and

motivates young adults. A broad exposure to different points of view and activities both affords students

high quality information and trains them to be active members of the citizenry (Delli Carpini and Keeter,

1996). Other research has shown that there is a dramatic increase in participation levels from before

college to after college, to a level that remains comparatively stable through adulthood. During the

college years, individuals become as attentive as they will ever be. Being exposed to the vigorous

collegial environment during the impressionable years is an advantageous coincidence that facilitates the

educational process. In contrast, those without a college education typically show a gradual, steady

increase in level of participation as they age. The intense exposure during years of high awareness for

4

those who graduated from college exacerbates a gap in participation most salient during the early

adulthood years (Strate et al., 1989; Converse and Niemi, 1971). But considering that all individuals

receive the benefits of aging, the participation gap increases, for college graduates are the sole recipients

of the interactive effects of aging and having acquired a higher level of education. As time passes on,

motivations and the exposure to different views and activities during the impressionable years serve to

generate the interest and efficacy that bridges education and later participation, a subjective mediation

process misguidedly attributed to the direct effect of education on participation in regression analyses.

There are other ways in which education may facilitate political participation. Education

contributes to the acquisition of communication skills, increases the ability to process and incorporate

higher quantities and more complex information, indoctrinates individuals into the process of learning and

self-teaching, provides better reading skills and the ability to identify useful information, and provides a

knowledge base permitting more coherent elaborations about new information and events (Bandura, 2002;

Dee, 2004; Delli Carpini and Keeter, 1996; Eveland, 1997, 1998; Eveland and Scheufele, 2000; Milligan,

Moretti, and Oreopolous, 2004). Education may develop cognitive skills that can nurture an ideological

identity (Krosnick, 1990; Fiske et al., 1990; see also Coté, 1996). Receiving a college education not only

exposes the individual to a stronger flow of information, but also encourages the development of

cognitive tools that facilitate the incorporation of political information into the schemata of the individual,

from which it becomes more accessible and prone to influence behavior. Throughout adulthood the

individual continues to cultivate her political knowledge commonly nurtured through media exposure.

This lifelong intellectual exercise may be entrenched in adults’ political engagement, which helps to

explain how college education affects later participation.

In summary, our simplifying idea about how education affects participation is the impressionable

years hypothesis. In this view, each generation is politically socialized principally during the

5

“impressionable years” of adolescence and early adulthood. Those years occur within a specific historical

period, whose norms, activities, and attitudes toward political participation in general are conveyed

through the individual’s educational experiences. The environment that surrounds an individual during

the impressionable years defines the types and the flows of information the individual is exposed to and

accumulates. It also may influence the interest and efficacy the individual may have toward it. We

describe the subjective residues of the combination of education and political socialization during the

critical period of young adulthood as “political engagement”—a general orientation that includes interest

in and attentiveness to politics. That youthful political engagement is presumed to persist into adulthood

and affect adult behavior. From this perspective, then, education and parental political socialization

influence the young person’s political engagement, which in turn has lasting effects on adult political

participation.

The Impressionable Years Hypothesis

The Impressionable Years hypothesis suggests that individuals are particularly susceptible to influence

during their late adolescence and young adult years, and that the predispositions and attitudes acquired

during these years endure throughout adulthood (Greenstein, 1965; Sears, 1975, 1981; Sears and Funk,

1999; Jennings and Stoker, 2005). High levels of sensitivity and the new political awareness experienced

during the preadult years, mainly nourished within the family environment, shape initial socio-political

development (Sears, 1990; Valentino and Sears, 1998).

This period of susceptibility to the formation and change of socio-political orientations is referred to

as the impressionable years. During the “impressionable years,” political orientations presumably

crystallize, thus influencing individuals for the rest of their lives. An important implication of this

perspective is that, under similar environmental stimuli, members of a same birth cohort share similar

socio-political orientations (Mannheim, 1928, 1952). The historical atmosphere in which peoples’ early

6

lives are embedded, especially when exposed to highly salient political events such as wars, crises, and

scandals are powerful determinants of the socio-political orientations they will hold through later

adulthood (Miller, 1992; Sears, 1990; Markus, 1979). But not all individuals will have the same early

experiences. Mannheim’s (1928, 1952) concept of a “generational unit” suggests that individuals located

in the same social strata should have similar experiences, and show generational commonalities in

adulthood (Cutler, 1974; Krosnick and Alwin, 1989; Miller and Sears, 1986).

The Impressionable Years hypothesis has two testable implications, then: (1) The persistence and

gradually increasing strength throughout adulthood of the political orientations acquired during the

preadult years, and (2) The particular sensitivity to political socialization of individuals during their young

adulthood in comparison to their later resistance to change as adults. These qualities of early

susceptibility and later persistence have been particularly enlightening in the formation and crystallization

of political attitudes under the stimulus of high flows of information proper of salient political events.

Family longitudinal studies have found that socio-political attitudes are less stable in the young offspring

than in their middle-aged parents; attitudes that are exposed to strong information flows tend to crystallize

in a more consistent fashion and become stronger as the individual ages (c.f. Jennings and Markus, 1977;

Sears and Valentino, 1997; Valentino and Sears, 1998). Considering the unparallel exposure and

sensitivity of the college years, education both crystallizes attitudes and trains individuals in participation,

a correspondence that becomes mediated as students leave behind the collegial environment.

Political Participation and Education, and the Impressionable Years

We argue that the cognitive skills of individuals who live in democratic systems are framed within the

rules and norms that shape the behaviors that are more valuable to their society, for instance, political

participation. This is especially true among those individuals who go through college education, a

process that takes place during the impressionable years for the great majority of the citizenry. Not

7

surprisingly, we expect that the process of education and the cognitive engagement that results from it

mediate the translation of political attitudes acquired through parental socialization into later political

participation.

But education does not happen in isolation. The college experience (from topics covered in

seminars, to the discourse recurrent in collegial discussions and debates, and to the issues addressed by

interest groups and on-campus organizations) is entrenched in the specific historical period in which it is

delivered. If this is so, then changes in the political attitudes of the masses should occur through

generational replacement. Analysis of age cohorts across time revealed generational differences in their

political socialization, helping us to understand the paradox of simultaneous turnout decline and higher

educational attainment across time. College education has become increasingly important over time;

declines in turnout are particularly concentrated amongst individuals with lower levels of education, who

are increasingly less likely to be strong partisans, interested in politics, or followers of the media (c.f.

Miller, 1992). In 1960, for example, a college graduate was 50% more likely to vote than someone who

did not graduate from high school, but in 2000 a higher 25% of Americans have attended college at some

point without substantive translations into participation (Patterson, 2002). Thus the overall turnout

decline may be due to the generational replacement of pre- and New Deal voters with a less politically

engaged post-New Deal generation (Miller and Shanks, 1996; Miller, 1992). These findings suggest that

education may not be a direct responsible at producing participation, but that it works with the political

socialization of young adults during specific historical periods to produce lasting participation.

Two main efforts to explain mediation effects of education on participation are also in place. The

first is the relative education hypothesis, which states that the absolute attainment of education does not

have a direct impact over political participation, but that it works as a “third variable” that provides

reputation, wealth, access to networks and institutions as well as to a higher labor market strata and

8

employment opportunities, which are the true facilitators for political participation (Nie, Junn, and

Stehlik-Barry, 1996; Helliwell and Putnam, 1999/2007; Tenn, 2005). Thus political participation is a

function of social standing and not of educational attainment alone (Hillygus, 2005). A second approach

is the political meritocracy hypothesis, which posits that both education and political behavior are caused

by a third variable, intelligence. General intelligence promotes the necessary cognitive engagement that

leads to both higher educational attainment and to be more involved in the political process (Jencks et al.,

1973). However, Hillygus (2005) shows that the content of college education spurs political engagement

regardless of intelligence; even college graduates without a social science background vote at higher

levels than individuals with no higher education.

This evidence suggests that the effects of education on political participation are intimately related

to the effects of political socialization, which more often than not coincide with the educational processes

of the individual. As models for predicting political behavior become increasingly interactive, the

premises of the Impressionable Years hypothesis facilitate the study of the relationship between education

and the political environment during a crucial period of susceptibility. The commonly attributed direct

effects of education over participation may be the result of mediation processes usually ignored in

regression analyses. It is our view that education is the result of parental socialization and socio-

economic status, a process that coincide with the life years in which individuals are more impressionable

by contextual stimuli thus imprinting the subjective political engagement necessary to produce a more

stable participation pattern through adulthood.

Data and Method

This study uses the Youth Parent Socialization Panel Study to explore how education and parental

political socialization facilitate political participation via political engagement. The data are longitudinal

and span from 1965 to 1997. The data are compounded by four waves: 1965, 1973, 1982, and 1997. The

9

first wave involved a probability sample of about 1,700 high school seniors (Generation 2), and 1,600 of

their parents (Generation 1) (Beck and Jennings, 1991). Jennings and Markus (1984) find the retention

rates of this study to be similar to other panel studies. There is a 56% response rate from 1965 to 1997 for

a total of 935 participants (Jennings, 2001). Tests reveal that there are no significant differences between

those individuals who dropped out and those who did not, thus our analyses make use of complete data

cases.

In order to represent the impressionable years, data from the parental and political social

environment are taken from the 1965 dataset. In three successive models, we test the predictability of

later political participation in 1973, 1982, and in 1997 based on the 1965 data. Specifically, our general

model is first fitted and then we conduct two different types of analyses to address whether mediation is

occurring or not and to what extent (Frazier, Tix, and Barron, 2004).

We implement SEM to examine the nature of the multivariate interactions, namely the latent

structures that are associated with several types of political participation. We develop a structural

equation model in which parental political socialization and education do not have a direct contribution on

political participation but rather they serve as catalysts for political engagement, which eventually leads to

political participation. In particular, we find that education and parental political socialization, both

variables widely assumed to have direct effects on political participation, are actually mediated through

political engagement in order to impact participation. Our aim is to propose an illustration of not only

what leads to political participation, but also how the environment that surrounds an individual during the

impressionable years, the experience of receiving a college education, and political engagement lead to

political participation.

SEM is particularly advantageous for our purposes because mediation for both education and

parental socialization can be adequately examined, in addition to controlling for measurement error. Two

10

other considerations are worth noting: (1) We are less interested in the amount of the variance in political

participation accounted for than in organizing its conceptual structure in terms of latent factors and their

relationships over time, and (2) nor are we focused on which latent factor is more or less important for

political participation, which has been already addressed by multivariate regression models, but on the

structural composition of the factors that lead individuals to participate.1

[Figure 1 about here]

Figure 1 illustrates our general structural equation model. The top two factors (SES and Parent

Engagement) are based on the 1965 data, representing the environment of the 17-18 year olds (Generation

2) who compound the units of analysis under observation. SES is taken from the family (i.e. Generation

1) and it represents the extent to which a family has provided the offspring with an enriched environment.

Parent engagement2 is a mixture of political participation and cognitive indicators that assess the parental

general orientation toward politics when their offspring were 17-18 years of age.3 The offspring’s

education, political engagement, and political participation are operationalized from 1973 onwards,

depending on the model in question. The offspring political engagement variables are similar to the

parent engagement variables, with the exception that offspring political participation is our outcome of

interest.

Operationalization

The data for the parents (Generation 1) are taken from the 1965 wave and are identical for all models.

Only one time point is used because the primary constructs of interest (family SES and parental

socialization) are assumed, in tandem with socialization theory, to be most relevant during the early stages

1 The relationship between latent factors is expressed in standardized parameters provided in the tables 1 through 3. Factor loads of structural components are also reported in these tables. 2 Taken from one parent due to data limitations. 3 We included parent political activity in a general latent variable for parent engagement under the assumption that parents’ interest, attention to, and participation in politics are holistically involved in the political socialization of the child.

11

of an individual’s lifespan.

Family SES is operationalized by four separate variables: (1) Duncan decile score for the

occupation of the household’s head, (2) Anticipated family income, (3) Highest level of education (in

years) of the household’s head, and (4) Spouse highest level of education (in years).4

[Table 1 about here]

Parent engagement is constructed through seven variables that account for political activities and

cognitive orientations toward politics. Newspaper reading is measured through a 4-value scale which

ranged from “never” to “daily”. Political magazine readership is dichotomous. Interest (i.e. paying

attention to political affairs) is measured by a 4-value scale ranging from “hardly at all” to “most of the

time”. Political participation is measured on a 4-value scale ranging from no activity (including no

voting) to high activity (including voting and two or more other political acts). Internal and external

political efficacies are assessed by a 3-value scale ranging from low to high. Political knowledge is

assessed by the summation of correct answers given to six questions on political knowledge.

For our first model Generation 2 (offspring) data are from the second (1973) wave. This is to

structure the extent to which the young-adult environment captured in the first (1965) wave influences

subjects when they were old enough to vote or consider the larger realm of politics out of high school.

Education is measured by years of college completed. Engagement variables are the same as parental

engagement variables with the exception of ‘radio’, which for Generation 2 fits well with the factor

structure. The dependent variable, political participation, is the sum of thirteen variables of political

activities including: Influencing others’ votes, attending political meetings, wearing a campaign button,

donating money to a party, contacting public officials, writing political letters to an editor, attending a 4 Cases in which the head of the household was a single mother (and the father was absent) the mother was named “head of household” for the first three variables and the spouse’s education was given a value of 0.

12

demonstration, participating in community affairs, voting in 1972 for president and congress, voting in the

1968 presidential election, and voting in 1970 for congress among other electoral activities.

Model Estimation for 1973 (Model 1)

Our models are run using maximum likelihood in EQS. The missing data are estimated through

maximum likelihood, as well. Standard errors are calculated using the Fischer method; robust methods are

used to correct the standard errors due to non-normality. Table 1 presents fitting statistics for the

hypothesized models and their respective standardized coefficients (factor loads). For Model 1 the

normalized estimate of multivariate kurtosis is not high enough to indicate significant non-normality

(2.5); however, robust methods are used. This structural model designs a direct effect from the

offspring’s political engagement latent, while all the other variables have only indirect effects over

political participation. All parameters are statistically significant at the .05 level using the robust standard

errors. Despite the high χ2 the model fit is highly satisfactory. No direct effects for parental SES and

engagement or offspring’s education are at work. Model 1 suggests that political participation may be

better understood through complex dynamics different than independent impacts attributed to explanatory

variables. Mediation roles such as that of the individual’s political engagement as constructed by earlier

family environments and the individual’s level of education will drive our attention below.

Model Estimation for 1982 (Model 2)

Model 2 uses the Generation 1 data from 1965 for family SES and parental engagement, but

replaces the Generation 2, 1973 data used in Model 1 with the 1982 data.5 This model shows somewhat

non-normal distributions (a multivariate kurtosis normalized estimate of 3.5). The fit indices are nearly

identical to those of Model 1 (Table 1). The standardized parameters indicate that the factor loadings are

also nearly identical. The only parameter that appears to be substantially different is the causal path 5 The 1982 education variable takes the number of college years reported in 1973 and adds the correspondent amount of years for the extra degrees or education acquired between 1973 and 1982. This same procedure is implemented in following estimations.

13

between parental engagement and offspring engagement—the parameter drops from .32 to .24. However,

offspring engagement still shows the same causal relationship with political participation. Overall, the

1973 model could only account for slightly more variance in political participation than the 1982 model

(43% compared to 39%, see Table 1).

Model Estimation for 1997 (Model 3)

Similar to the 1982 model, this model replaces education, activity, and political engagement latent

factors with the 1997 data. The method of model estimation was the same as the 1973 and 1982 models.

All standardized parameters are statistically significant.

A comparative analysis of the three models highlights some considerations worth mentioning. It

is noticeable that factor loadings in our general model are fairly stable across time with the exception of

low increases detectable in offspring political engagement scales; specifically, the factor loadings for

external efficacy and political interest (see Table 1). Also, there is a substantial increase in the factor

loadings in newspaper reading and political radio usage. This increase means that, as the subjects

approached middle age, the latent factors became more consistent, probably due to the gain in maturity

and life experience that comes with aging.6

The influence of college education over the political engagement of Generation 2 drops from .42

to .40 and from .40 to .38, respectively. Despite there is a pattern of decline across time, these differences

are trivial. Alternatively, we interpret these values as evidence of college educations’ impact stability.

Even when this education was achieved during the subjects’ early twenties, it still plays a relevant role in

their political engagement decades after.

The impact of parental political engagement on the political engagement of Generation 2

decreases from .31 to .24 to .26. Despite this somewhat larger drop, this finding exemplifies how parental 6 This does not suggest that there were mean increases in the variables themselves. In fact, political interest and external efficacy show similar means in 1973 and 1997 while newspaper readership and radio usage actually dropped from 1973 to 1997.

14

political socialization during youth has a lasting effect on offspring’s political engagement, even well into

their forties. It is not surprising that this effect diminishes given the variety of other sources of

socialization out of the family environment that influence political engagement during adulthood, most

likely after parents have passed away or after the offspring leaves the family nest.

Lastly, and most important, the effect of Generation 2 political engagement on political

participation is very strong and remains so over time. This finding is reliable and all of our models

[Table 2 about here]

continued to fit convincingly well despite college education and parental political socialization did not

hold a direct influence over political participation throughout the time span of our analysis. By always

being treated as indirect actors and not as direct causal forces of political participation, the notion of

political engagement as a mediator between these well-established predictors and political participation is

supported.

Political Engagement as a Mediator

Part of our argument is based on the advantages that SEM can bring to the exploration of mediation

dynamics in comparison to multivariate regression models. On these grounds, our structural models show

that the direct impacts commonly attributed to college education and parental socialization over political

participation are actually mediated through political engagement. However, an essential question for the

validity of our inferences is if whether or not absent direct effects should be included in our models. We

thus can implement Wald tests to check on this alternative hypothesis, and the Wald tests will show the

extent to which the fit of the model worsens as parameters included in the model are dropped out. This

can be assessed by the behavior of fit statistics; for instance, how much does the χ2 increases as

parameters of interest are left out of the structural model. An alternative and more straightforward way to

address our question, however, is to insert causal paths from education and parental socialization toward

15

participation (see dashed paths in Figure 1). By including these direct effects we can use Wald tests to

assess the relevance they play and, therefore, contrast our findings with the bulk of evidence produced

through multivariate regression models.

By testing both the direct and indirect effects we are allowed to draw the contextual and

psychological paths individuals experience as they move on in their daily lives. By analyzing the

complexity of the structural dynamics associated to political participation across time, we are also able to

distance ourselves from the crude way of interpreting individuals’ actions as the direct result of variables

such as parental SES. More centered in the holistic process of how variables interact to produce a specific

outcome, this approach also helps to elucidate the internal mechanisms through which well-established

political participation paradigms (e.g. the SES model and the parental socialization model) come to

fruition.

Does college education substantially influence political participation via political engagement, or

is there a story of class differences in which the SES of the family has a direct impact on offspring’s

political participation? The hypothesized model is presented in Figure 1, in which the dashed paths

represent the direct effects. This model is structurally identical to our main model with the exception that

direct causal paths from education, and parental SES and engagement toward participation have been

included. Explicitly, this hypothesized model tests if, for example, parental SES has a meaningful direct

effect on political participation in addition to the indirect effect in which it mainly serves to facilitate the

college education of the offspring, which in turn facilitates political engagement.

None of the parameters for the new direct effect paths are statistically significant. This finding

holds for all of the three waves. The only exception is the path from college education in 1997, which has

a trivial value of –.09. The lack of a strong, positive direct effect over participation is probably due to the

powerful path of mediation. A clearer picture can be drawn from our Wald tests results. These Wald tests

16

are conducted for each of the three waves (see Table 2). In every case, the included paths from education,

and parental SES and engagement show miniscule increments in the χ2 statistic. The lack of significance

in all these tests indicate that dropping the direct effect paths does not affect the overall fit of the model;

concisely, the paths of direct effect are not essential to the model. These results provide evidence in favor

of the mediation hypothesis, and equally important, they also afford us the foundations to test for

additional dynamics that may be driving our main findings.

[Figure 2 about here]

Deconstructing the Relative Effects of Education and SES

Another possible critique to the previously presented models is that education is given a primary role that

should not have. It may be the case, for example, that education effects are really masking a deeper story

about SES in that more privileged families are the source of both highly educated and more politically

active offspring. Thus, the view that the effect of parental SES over offspring political engagement is

mediated through college education may be erroneous. Furthermore, not only parental SES but also the

offspring’s composite SES might be more relevant factors to political engagement than college education

alone.

We thus test for these notions by proposing a model that allows for direct effects from parental

SES, offspring college education, and offspring income over offspring political engagement. This first

alternative model (Model 4) uses data from the 1973 wave. Subsequent models follow the format of our

original models and indicator variables have been removed for the sake of visual simplicity (Figure 2).

By allowing a direct effect from parental SES to offspring political engagement we are able to explore,

firstly, the relevance of college education as a direct versus a mediator actor, and secondly, to what extent

are offspring political engagement and consequently political participation better explained by the SES

model. The alternative explanation would be that parental SES mainly facilitates college education,

17

which in turn would be the most important factor with respect to political engagement, as originally

hypothesized.

Model 4 shows that the direct path from parental SES to offspring political engagement is not

statistically significant (Table 3). This provides support for the mediation of parental SES through

college education onto offspring’s political engagement. Additionally, offspring income, at least by 1973

(when the offspring is about 26 years of age) is irrelevant in the direct prediction of political engagement.

[Table 3 about here]

Wald tests corroborate that the non-significant paths can be dropped from the model. This finding offers

further support for the notion that parental SES is mostly beneficial for offspring political engagement and

political participation in that it allows for college education.

The model for the 1982 wave (Model 5) shows a slightly different situation. The direct effect

from parental political engagement losses statistical significance while the direct effect from parental SES

to offspring political engagement still does not show statistical significance (Table 3). However, by the

time the offspring reached about 35 years of age, their own income has become a statistically significant

predictor of political engagement. Wald tests on our alternative explanation, nevertheless, highlight that

the comparative importance of college education as a mediator cannot be ignored: Dropping the income to

political engagement path costs only 17 χ2 units to the fit quality of the model, while dropping the parental

SES to college education and the subsequent college education to political engagement paths signifies 35

and 122 χ2 units to the fit quality of the model, respectively. This finding shows some support for the SES

model as it is known today, but more importantly, it provides further evidence for the comparative

explanatory power advantage of mediation dynamics, especially as life-related circumstances of

individuals change throughout the aging process.

Running the same model for the 1997 wave (Model 6) evinces that parental SES still does not

18

directly predict offspring political engagement. Interestingly, these findings bring to light that the

parental SES effect over offspring political engagement and participation is eminently mediated

throughout a three-decade life span. Findings also show that now that the respondents are about 51 years

of age, income became more tied to engagement. At middle age, parental SES is now predicting offspring

income; however, Wald tests show that excluding this path form the model only adds 8 χ2 units to the fit

quality of the model. Additionally, offspring income predicts political engagement, but not as strongly as

[Table 4 about here]

college education. While dropping the income to engagement path adds 33 χ2 units, dropping the

education to engagement path adds 75 χ2 units to the fit quality of the model. These findings portray a

clearer picture of the complex interactions between parental socialization and offspring socio-economic

variables throughout a good portion of individuals’ lifespan.

Results across all the three waves bring support to our original structural model. Political

engagement—the main mediator toward political participation—is not the direct result of parental SES

but mainly of mediation, especially through college education. Additionally, college education showed to

be a powerful and stable mediator for political engagement, even when compared to income and

throughout a good set of alternative models. But this finding has a further implication: It indicates rather

clearly that the SES model as it is mostly known today—one in which the political effects of education

are largely due to its relationship to class—needs further specification. Class is primarily important in

that it facilitates offspring’s access to higher education, which in turn nourishes the rich environment

needed to develop political engagement. By itself, parental SES has no direct impact over either offspring

political engagement or participation, while education is consistently connected to parental SES and has a

large impact over offspring political engagement.

An Alternative Model

19

Most of the literature on political participation is divided between electoral and non-electoral

participation. Despite both types of participation are behavioral phenomena their differences may be

based on the costs, resources, and attitudinal dispositions proper to each type of participation. Our

dependent variable, however, is designed to capture a broad configuration of voluntary actions imparted

by individuals in the search of a political outcome at different levels of the political system. It is not our

perspective that structural models specified on the basis of type of participation should be substantially

different, especially when considering that participation can be an “orientation” just like attitudes. As a

test for robustness we also consider the possibility that merging both types of participation in a single

scale of political activity might be perceived as inadequate. Accordingly, we disaggregated our dependent

variable, and the original hypothesized structural models were re-ran by using all the types of non-

electoral political participation items alone as our outcome of interest.

Table 4 presents the fitting statistics of this alternative model. Findings show that the fit indices

are nearly identical to our original structural model. The only difference is that less variance in political

participation is explained (39 to 43% for the original model, and 32 to 35% for the alternative model).7

These differences are apparently trivial for the purposes of this paper.

Final Remarks and Conclusions

This piece has presented evidence that the main assumption of the Impressionable Years hypothesis—that

significant events during young-adulthood serve as a political socialization mechanism that have lasting

effects on attitudes and predispositions—can in fact be extended to actual political behavior. We have

demonstrated that an interaction between offspring college education and parental SES and engagement

forge a broad offspring engagement orientation toward politics which in turn predicts offspring political

7 Appendices A through C present a summary of additional robustness tests that were developed on the basis of alternative hypotheses and on the basis of possible additional variables not included in our latent factors. These appendices also contain tests for assumptions and respective model estimations, as well as further specifications on the construction and the general stability of our structural model.

20

participation. The value of this piece is thus its general illustration of how the political process happens at

the individual level. This piece brings structure to individuals’ political processes through the theoretical

organization of commonly isolated factors into latent constructs with causal mapping.

Beck and Jennings (1991) exemplify the transmission of attitudes from parent to offspring and

how this correlation changes over time. But the direct relationship of specific attitudes between parent

and offspring is not the only interesting piece of information in socialization processes. As the authors

write, “the parental legacy is to be found in similarities of broad orientations toward politics rather than

the one-to-one correspondence between particular parental and offspring traits” (p. 745, emphasis added).

Accordingly, this piece is not interested in the actual correspondence between parent and offspring

attitudes; rather it is focused in how the most relevant aspects of the family environment interact

throughout time with the later endowment of the offspring to generate a ‘broad orientation toward

politics’ (which we called ‘engagement’) and test its influence over political participation.

Beck and Jennings (1991) also estimate the bivariate correlation between parents and offspring on

their measure of engagement (termed ‘politicization’) and find a consistent correlation. This bivariate

similarity lent credence to the idea that engaged families can produce engaged offspring. However, it is

most certainly not the case that offspring are political carbon copies of their parents. Parent/offspring

correlations on different traits do erode over time. This call attention to the importance of post-

adolescence socialization factors, most likely college education and period effects which may be

disproportionately affecting younger cohorts in the general electorate, reiterating the notion that any one

of these factors cannot independently account for political participation. Apart from this observation, the

fact that parental engagement, as measured when their offspring were about 18 years of age, continues to

influence offspring’s political engagement 8, 17, and even 32 years afterwards is remarkable.

Of course, this research is not without its faults. In particular, the longitudinal dataset used herein

21

is one of very few of its kind that follows the same cohort over such an impressive span of time.

However, the youth in this panel came of age during a particularly turbulent epoch, with the controversies

of the Vietnam War and Watergate being powerfully salient historical events. A consideration of both

impressionable years and generational effects would find this relevant. As a result of these events, for

example, the youth panel is far less committed to the political parties than their parents (and has remained

less committed for decades) indicating a generational difference (Jennings and Markus, 1984). Given this

disparity, it is not surprising that the bivariate correlations between parent and offspring attitudes are

modest at best, and weak by the time the offspring reached adulthood. It is worth noting, however, that

different types of similarity tests between parent and offspring answer different questions. The direct

correlation between parent and offspring on any given attitude can be without doubt attenuated by the

historical context, which affects the offspring in a disproportionate manner. A more general test, and the

one explored in this piece, is to examine the similarity between parental political engagement and that of

the offspring.

This piece is also consistent with the optimistic view that education can facilitate political

participation. Findings show that this is true but only inasmuch as education facilitates political

engagement. As elections in the last half of the 20th century indicates, it is clearly not the case that a more

educated public is necessarily more inclined to turnout, or actually toward many other types of

participation. The other active ingredients needed for political participation are, therefore, related to the

environmental contributors to political engagement. Our structural model suggests, in accordance with

the Impressionable Years hypothesis, that the timing of the interaction between relevant events, the family

socialization environment, and the political development of the individual deserves much closer attention.

College education, which occurs in a politically affluent environment, is likely to have long lasting effects

because of its timing during the impressionable years period.

22

Findings also suggest that the paradox of declining participation while educational attainment has

increased may be an imprecision of the research produced through regression analysis, which stipulates a

direct effect of education over participation. The reliability of this relationship (most of it re-produced by

the same methodological approach) may have conditioned much of our theories of political behavior. Our

structural model suggests that this paradox may be the byproduct of a spurious correlation induced by

regression modeling, for the impact of education over political participation is indirect, mediated through

political engagement. Our findings do not suggest, of course, that research produced through regression

analysis is at fault, but that there are many risks involved in the development of a discipline when it over-

relies on a single methodological approach.

We do not attest that Structural Equation Modeling is a more adequate method for the analysis of

political participation or any other political phenomena, but that in this particular case, it offers an

improvement over, or it is at least an alternative to, an endemic regression typology in which theory-

driven assumptions about complex causation, mediation and endogenous effects cannot be readily tested.

Research in the social sciences need also to develop models that explain how variables are related and in

what sense constructs can be ordered and projected, rather than only focusing on the fact that they are

related. Questions of what and how are not mutually exclusive in scientific research, but to the contrary,

they can bring great benefit to the development of knowledge if analyzed in conjunction. We also believe

that, by increasing the methodological repertoire within the discipline, it would not only open a better

understanding of, and communication paths with, other disciplines but also would help to incorporate

more efficiently other lines of research within the discipline that gauge their research questions from

different methodological perspectives. This, in return, will help to put findings into a more diverse

perspective and help to avoid the development of close-ended theories constrained by methodological

embedment.

23

Finally, we believe this piece offers a great variety of implications that could be applied to

numerous political contexts. The high profile of recent presidential elections, for example, involves

college campuses with very high levels of interest in the campaign. The presence of such interest sparks

discussions, debates, and most probably increases consumption of political information, making the

college atmosphere an ideal environment for political learning. It is likely that such intense periods in

politics have lasting effects on the youth generations, and more surely as the level of political fervor

across parties and ideologies becomes a persistent stimuli. The effects of these educational atmospheres

of mobilization are to be tempered by the youth’s family socialization and then crystallized into more

permanent political attitudes that in turn will sum up to their general political engagement. The eventual

political participation of the youth would project and reaffirm their young-adulthood experiences

throughout the rest of their lives with relative stability thus propagating onto the future the environment in

which they became politically engaged individuals.

24

Tables and Figures

Table 1. Statistics and Coefficients of General Structural Equation Models for 1973, 1982, and 1997*

Path** Coeff. Model 1 (1973) Coeff. Model 2 (1982) Coeff. Model 3 (1997)

(Parent SES, Parent Eng) .75 .75 .75 (Child Eng, Parent Eng) .31 .24 .26

(Knowledge, Parent Eng) .66 .66 .65 (Ext. Eff., Parent Eng) .40 .39 .39 (Int. Eff., Parent Eng) .56 .56 .56 (Activity, Parent Eng) .54 .54 .54 (Interest, Parent Eng) .59 .59 .60

(Magazines, Parent Eng) .51 .52 .51 (Newspaper, Parent Eng) .53 .53 .53

(Education, SES) .49 .48 .46 (Duncan, SES) .71 .71 .70 (Income, SES) .70 .69 .69

(Head Education, SES) .73 .74 .74 (Wife Education, SES) .65 .65 .65

(Child Eng., Education) .42 .40 .38 (Pol. Activity, Child Eng) .65 .62 .65

(Knowledge, Child Eng) .63 .61 .61 (Ext. Eff., Child Eng) .21 .36 .34 (Int. Eff., Child Eng) .47 .54 .49 (Interest, Child Eng) .46 .59 .57

(Magazines, Child Eng) .44 .44 .42 (Newspaper, Child Eng) .36 .43 .56

(Radio, Child Eng) .15 .15 .37 d.f. 167 149 167

N 636 636 636 Chi-square (χ 2) 387 365 380

NFI .87 .88 .87 CFI .92 .93 .93

RMSEA .05 .05 .05 R-square (R2) .43 .39 .43

* All coefficients in Models 1, 2, and 3 (1973, 1982, and 1997 respectively) are standardized and statistically significant (p<.05). All models include maximum likelihood estimation for missing data. ** The depiction of directional paths are (to, from); e.g. the path that goes from Child Political Engagement to Political Activity is depicted as (Pol. Activity, Child Eng).

25

Table 2. Wald Tests for Direct Effects of Three Common Predictors on Political Activity (Dashed Paths in Figure 1)

Coefficient χ2 Improvement p-value Model 1 (1973)

Parent SES -.04 .3 .54 Parent Engagement .11 1 .39

Education .03 3 .55 Model 2 (1982)

Parent SES -.09 2 .49 Parent Engagement .05 1 .60

Education .03 .3 .56 Model 2 (1982)

Parent SES -.04 1 .58 Parent Engagement -.01 .01 .89

Education -.09 5 .15

Table 3. Coefficients of Three Alternative Structural Equation Models for 1973, 1982, and 1997*

Path Coeff. Model 4 (1973) Coeff. Model 5 (1982) Coeff. Model 6 (1997)

(Parent SES, Parent Eng) .74 .74 .74 (Child Eng, Parent Eng) .29 n.s.** .25 (Education, Parent SES) .54 .51 .43

(Income, Parent SES) n.s. n.s .15 (Child Eng., Parent SES) n.s. n.s. n.s.

(Child Eng., Income) n.s. .24 .26 (Pol. Activity, Child Eng) .62 .54 .65 * Reported coefficients are statistically significant (p<.05). Coefficients of indicator variables not included. ** Non-statistically significant.

Table 4. Fit Statistics for General Structural Equation Models for 1973, 1982, and 1997 (Political Activity Not Including Voting)

d.f. N χ2 NFI CFI RMSEA R2

Model 1A (1973) 149 636 359 .86 .93 .05 .35 Model 2A (1982) 149 636 360 .88 .93 .05 .32 Model 3A (1997) 167 636 370 .88 .93 .04 .33

26

Figure 1. General Structural Equation Models for 1973, 1982, and 1997*

* SES [of parents] and Parent Engagement data are from 1965. The variables Education, Child Political Engagement, and Political Activity belong to the offspring. Results for these models appear in Table 1. Dashed paths correspond to Wald tests for direct effects (results of these tests appear in Table 2).

27

Figure 2. Alternative Structural Equation Models for 1973, 1982, and 1997*

* Results for these models appear in Table 3.

28

Appendix A

Model Construction: Generation 1 (Parent) Data

Two options were considered for parent political engagement: a one factor model, and a two

factor model which broke down “political activities” and “self-schema” into two related factors

(see Figures 3 and 4).

Testing Assumptions

Analysis of normality of the data was conducted. No univariate outliers were detected by visual

inspection of histograms. Uni- and multivariate normality was assessed within the model runs in

EQS. In all three models, the normalized estimate of Mardia’s Coefficient is fairly high (7.8,

7.8, 6.9, respectively). In all three models, the same 5 cases were named as the largest

contributions to normalized multivariate kurtosis; however further inspection of these cases did

not reveal any apparent abnormalities. Thus, these cases were included in data analysis.

Because of the non-normality in the data, robust estimation methods were used.

Model Estimation

All models were run analyzing the covariance matrix using maximum likelihood and correcting

for non-normality by using robust statistics. Two different models were run in order to test if

parental political actions and activities (like media usage, and voting) should be considered a

separate factor from more cognitive variables, such as political interest, efficacy, and knowledge.

Model A1 considered only two factors, SES and Political Engagement, which are allowed to

correlate. Although the χ2 for the model is fairly high, in general the fit indices were fairly good

(see Table 5). The model with standardized parameters is presented in Figure A1. All

parameters are statistically significant, using the Fisher method of computing robust standard

errors.

29

Model A2 is presented as an alternative model, with three factors: SES, Political

Activities, and cognitive variables. Model A2 and its standardized parameters appear in Figure

A2. As indicated in Table 5, a 3-factor model does not present a significant increase in model

fit. Analysis of the model indicates a very high correlation between the Political Activities and

Self-Schema latent variables. For these reasons, a 2-factor model was settled on for parental

data, for the sake of avoiding unnecessary complexity.

Because 151 cases are lost to listwise deletion in these models, Model A2 was re-ran

using the maximum likelihood estimator for missing data. This produces identical fit indices and

similar parameter estimates.

Figure 3. Parent Data Measurement Model (Model A1—Two Factor Model)

30

Figure 4. Parent Data Measurement Model (Model A2—Three Factor Model)

Table 5. Fit Statistics for Parent Data Measurement Models

d.f. N χ2 NFI CFI RMSEA Model A1: Two Factor Model 64 747 218 .91 .93 .06 Model A2: Three Factor Model 62 747 214 .91 .93 .06 Model A3: Two Factor * 64 898 269 .91 .93 .06 * Includes maximum likelihood estimation for missing data

31

Appendix B Model Specification for Generation 2 Data

As with the Generation 1 data, analyses were carried out to see if the “political engagement”

latent variable should actually be considered two separate latent variables. However, analyses

indicated that a 2 factor model for Political Engagement could not be identified; thus Political

Engagement was kept as one factor.

Model B1, which is presented in Figure B1, demonstrates the test of factor structure with

Political Engagement, and the measured variables “Education” and “Activity.”

Testing Assumptions

No outliers were detected. The normalized estimate of Mardia’s Coefficient was -2.8, indicating

that the data are not substantially non-normal at the multivariate level; however, since the parent

data (which will eventually be added to the model are), all subsequent models were run using

maximum likelihood and robust statistics.

Model Estimation

The fit indices for Model B1 are presented in Table B1. While the fit is somewhat adequate, one

parameter emerges as somewhat unusual. The factor loading for watching political TV is

substantially lower than all other factor loadings. Other research (Murugesan, 2007) indicates

that despite the fact that TV news/politics-watching is correlated with other forms of media

usage, it is less highly correlated, and might be a different type of media usage because of its

passive nature. Because of this, the measurement model was re-ran eliminating TV from the

dataset. This resulted in similar parameters, but marginally improved fit indices (see Model B2,

Table B1).

32

Figure 5. Subject Data Measurement Model (Model B1—One Factor Model)

Table 5. Fit Statistics for Subject Data Measurement Models

d.f. N χ2 NFI CFI RMSEA Model B1: One Factor Model 34 501 117 .82 .86 .07 Model B2: One Factor Model (No TV) 26 501 78 .86 .91 .06

33

Appendix C La Grange Multiplier Tests for All Three Time points

A La Grange Multiplier test was run on the 1973 data, testing to see if additional parameters (any

relationships between factors, variables, or errors) would create a substantial improvement in

model fit. Given how large the model is in terms of variables, it is not surprising that close to

100 parameters come up as statistically significant, indicating that they would results in an

improvement in χ2. One cannot simply “go hunting” for significant parameters and add them to

the model in order to create a more desirable χ2. Analyses of such data should consider the size

of the effect and how much it makes sense with respect to the model. The top ten parameters (in

terms of χ2 gained by adding the parameter) for the 1973 data are listed in Table 6A. Each is

listed as “Y,X” such that the first variable is predicted, and the second is the predictor. In the

first case, parental political knowledge significantly predicts child political knowledge; this

indicates that, above and beyond the effect of the “Parent Engagement” factor on the “Subject

Engagement” factor, parent knowledge directly affects child knowledge. The second parameter

indicates that child education predicts parental political knowledge—a nonsensical finding. The

remainder of the other parameters indicate either 1) direct effects between variables within the

same factor, or 2) direct effects between a factor and a variable. For example, parental SES is

predicting parental interest in politics, above and beyond the correlation between SES and parent

engagement allowed for in the original model. More interesting are the education parameters:

both combinations of education and knowledge are significant, indicating that either variable

could predict the other. However, none of these parameters, with the largest contributions in

additional χ2 are the parameters of interest in terms of looking at the main DV: child political

participation.

Further down on the list are potential parameters, which, if added would contribute a

34

small amount of χ2. The first two indicate that either parental activity or external efficacy both

predict child activity later in life. The third and fifth parameters (child activity in 1973

predicting adult variables 1965) are nonsensical. The next two indicate parameters between

child activity and knowledge are significant in both causal directions.

Similarly, La Grange multiplier tests were run with the 1982 and 1997 data, with similar

results (see Tables 6B and 6C). Both the later dates do not show substantial improvements with

added causal paths from either variables or factors to child political participation. Most

relevantly, looking at the tests over time indicate that parental activity, political knowledge, and

interest are consistently variables that could added to have direct effects above and beyond the

impacts of their respective factors.

Table 6A. LaGrange Multiplier Test for 1973

Top 10 Added Parameters* χ2 Improvement Probability

(Knowledge, Parental Knowledge) 42 <.001 (Parental Knowledge, Education) 37 <.001

(Knowledge, Education) 32 <.001 (Interest, Newspaper) 28 <.001 (Newspaper, Interest) 24 <.001

(Parental Interest, Parental Newspaper) 21 <.001 (Parental Newspaper, Parental Interest) 20 <.001

(Parental Interest, SES) 18 <.001 (Education, Knowledge) 17 <.001

(Interest, Education) 16 <.001 Added Parameters Involving the DV

(Polit. Activity, Parental Pol. Activity) 14 <.001 (Polit. Activity, Parental Ext. Efficacy) 13 <.001 (Parental Pol. Activity, Polit. Activity) 12 <.001

(Knowledge, Polit. Activity) 8 .005 (Polit. Activity, Knowledge) 8 .006

(Parental Ext. Efficacy, Polit. Activity) 5 .029

* For simplicity parental variables are indicated while offspring variables are not, unless necessary. Thus, if the variable is stated as “Knowledge”, this means it is the offspring’s knowledge, otherwise it will be indicated as “Parental Knowledge.” This rule also applies to Tables 6B and 6C.

35

Table 6B. LaGrange Multiplier Test for 1982

Top 10 Added Parameters χ2 Improvement Probability

(Knowledge, Parental Knowledge) 46 <.001 (Parental Knowledge, Knowledge) 36 <.001

(Knowledge, SES) 27 <.001 (Knowledge, Education) 25 <.001

(Interest, Education) 25 <.001 (Knowledge, Parental Engagement) 24 <.001

(Parental Interest, Parental Newspaper) 20 <.001 (Parental Newspaper, Parental Interest) 20 <.001

(Knowledge, Head’s Education) 18 <.001 (Ext. Efficacy, Int. Efficacy) 18 <.001

Added Parameters Involving the DV

(Wife’s Education, Polit. Activity) 8 .003 (Polit. Activity, Wife’s Education) 5 .02

(Knowledge, Polit. Activity) 5 .03 (Polit. Activity, Knowledge) 5 .03

(Polit. Activity, Parental Polit. Activity) 4 .04 (Interest, Polit. Activity) 4 .05 (Polit. Activity, Interest) 4 .05

Table 6C. LaGrange Multiplier Test for 1997

Top 10 Added Parameters χ2 Improvement Probability

(Knowledge, Parental Knowledge) 41 <.001 (Knowledge, SES) 34 <.001

(Knowledge, Parental Engagement) 33 <.001 (Parental Knowledge, Knowledge) 25 <.001

(Knowledge, Head’s Education) 22 <.001 (Knowledge, Education) 22 <.001

(Knowledge, Duncan SES) 21 <.001 (Ext. Efficacy, Int. Efficacy) 21 <.001 (Int. Efficacy, Ext. Efficacy) 21 <.001

(Parental Interest, Parental Newspaper) 20 <.001 Added Parameters Involving the DV

(Parental Polit. Activity, Polit. Activity) 14 <.001 (Polit. Activity, Parental Polit. Activity) 8 .004 (Polit. Activity, Parental Ext. Efficacy) 6 .015

(Radio, Polit. Activity) 5 .02 (Polit. Activity, Radio) 5 .02

(Wife’s Education, Polit. Activity) 4 .04 (Interest, Polit. Activity) 4 .04 (Polit. Activity, Interest) 4 .04

36

References

Alwin, Duane Francis, Ronald Lee Cohen, and Theodore Mead Newcomb. 1991. Political Attitudes Over the Life Span. Madison, Wisconsin: The University of Wisconsin Press.

Beck, Paul Allen and Kent M. Jennings. 1991. “Family traditions, political periods, and the

development of partisan orientations.” Journal of Politics, 53(3), 742-763. Bennet, S. E. 1991. “Left behind: Exploring declining turnout among noncollege young whites,

1964-1988.” Social Science Quarterly, 72(2), 314-333. Bollen, K. A. 1989. Structural Equations with Latent Variables. New York: Wiley-

Interscience. Converse, Philip, E. 1972. “Change in the American Electorate.” In Angus Campbell and

Philip Converse (Eds.) The Human Meaning of Social Change. New York: Russell Sage Foundation.

Converse, Philip. E., and Niemi, R. 1971. “Non-voting among young adults in the United

States.” In Political parties and political behavior, ed. W. Crotty, D. Freeman, and D. Gatlin (pp. 443-466). Boston: Allyn and Bacon, Inc.

Delli Carpini, M. X., and Keeter, S. (1996). What Americans know about politics and why it

matters. New Haven: Yale University Press. Denny, Kevin, and Doyle, Orla. (2008). “Political interest, cognitive ability and personality:

Determinants of voter turnout in Britain.” British Journal of Political Science, 38, 291-310.

Frazier, Patricia A., Tix, Andrew P., and Barron, Kenneth E. (2004). Testing moderator and

mediator effects in counseling psychology research. Journal of Counseling Psychology, 51(1), 115-134.

Glenn, N. D. 1974. “Aging and Conservatism.” Annals of the American Academy of Political

and Social Science, 33: 176-186. Hillygus, D. S. (2005). The missing link: Exploring the relationship between higher education

and political engagement. Political Behavior, 27(1), 25-47. Jennings, Kent M. and Laura Stoker. 2005. Aging, Generations, and the Development of

Partisan Polarization in the United States. Santa Barbara, CA: Univ. of California Press. Kent M. Jennings, and Markus, G. B. 1977. The effect of military service on political attitudes:

A panel study. American Political Science Review, 71, 131-147. Kent M. Jennings, and Markus, G. B. 1984. “Partisan orientations over the long haul: Results

37

from the three-wave political socialization panel study.” The American Political Science Review, 78(4), 1000-1018.

Kent M. Jennings, and Niemi, R. G. (1968). The transmission of political values from parent to

child. The American Political Science Review, 62(1), 169-184. Kline, R. B. (2004). Principles and Practice of Structural Equation Modeling. New York:

Guilford Press. Krosnick, Jon A., and Duane F. Alwin. 1989. “Aging and Susceptibility to Attitude Change.”

Journal of Personality and Social Psychology, 57, 416-425. Krosnick, Jon A., and Duane F. Alwin. 1989. “Aging and Susceptibility to Attitude Change.”

Journal of Personality and Social Psychology, 57, 416-425. MacCallum, R. C. and Austin, J. T. 2000. “Applications of structural equation modeling in

psychological research.” Annual Review of Psychology, 51, 201-226. Markus, G. B. (1979). The political environment and the dynamics of public attitudes: A panel

study. American Journal of Political Science, 23(2), 338-359. McLoyd, Vonnie C. 1998. “Socioeconomic disadvantage and child development.” American

Psychologist, 53(2), 185-204. Miller, W. E. 1992. “The puzzle transformed: Explaining declining turnout.” Political

Behavior, 14(1), 1-26. Miller, W. E., and Shanks, J. M. 1996. The New American Voter. Cambridge: Harvard

University Press. Patterson, T. E. 2002. The Vanishing Voter. New York: Alfred A Knopf. Piaget, J. and A. M. Weil. 1951. “The Development in Children of the Idea of the Homeland

and of Relations with Other Countries.” International Social Science Bulletin, III. Rosenstone, S. J., and Hansen, J. M. 1993. Mobilization, participation, and democracy in

America. New York: Macmillan publishing company. Sears, David O. 1990. “Wither political socialization research? The question of persistence.” In

O. Ichilov (Ed.), Political Socialization, Citizenship, Education, and Democracy. (pp. 69-97). New York: Teachers College Press.

Sears, David O., and Funk, C. L. 1999. Evidence of the long-term persistence of adults’

political predispositions. The Journal of Politics, 61(1), 1-28.

38

Sears, David O., and Shery Levy. 2003. “Childhood and adult political development.” In D. O. Sears, L. Huddy, and R. Jervis (Eds.). Oxford Handbook of Political Psychology. (pp. 60-109), Oxford: Oxford University Press.

Sears, David O., and Nicholas A Valentino. 1997. “Politics matters: Political events as catalysts

for preadult socialization.” American Political Science Review, 91(1), 45-65. Sears, David O. and Carolyn L. Funk. 1999. "Evidence of the Long-Term Persistence of Adults'

Political Predispositions." Journal of Politics, 61, 1-28

Sears, David. O. 1975. “Political socialization.” In F. I. Greenstein and N. W. Polsby, Eds. Handbook of Political Science, Reading, MA: Addison-Wesley.

Sears, David. O. 1981. “Life stage effects on attitude change, especially among the elderly.” In

S. B. Kiesler, J. N. Morgan, and V. K. Oppenheimer (Eds.). Aging: Social change (pp. 183-204). New York: Academic Press.

Strate, J. M., Parrish, C. J., Elder, C. D., and Ford, C. (1989). Life span civic development and

voting participation. American Political Science Review, 83(2), 443-464). Turkheimer, Eric, Haley, Andreana , Waldron, Mary, D'Onofrio, Brian, and Gottesman, Irving I.

(2003). Socioeconomic status modifies heritability in young children. Psychological Science, 14(6), 623-628.

University Press. Valentino, N. A., and Sears, D. O. (1998). Event-driven political communication and the

preadult socialization of partisanship. Political Behavior, 20(2), 127-154. Winick, Myron, Meyer, Knarig K., and Harris, Ruth C. (1975). Malnutrition and environmental

enrichment by early adoption. Science, 190(4220), 1173-1175 Zaff, J. F., Malanchuk, O., Michelsen, E., and Eccles, J. (2003, March). Socializing youth for

citizenship. (Working paper no. 03). College Park, MD: Center for Information and Research on Civic Learning and Engagement.